Array
(
    [content] => 
    [params] => Array
        (
            [0] => /forum/index.php?threads/intel-is-trapped-flawed-vision-bad-execution.16419/
        )

    [addOns] => Array
        (
            [DL6/MLTP] => 13
            [Hampel/TimeZoneDebug] => 1000070
            [SV/ChangePostDate] => 2010200
            [SemiWiki/Newsletter] => 1000010
            [SemiWiki/WPMenu] => 1000010
            [SemiWiki/XPressExtend] => 1000010
            [ThemeHouse/XLink] => 1000970
            [ThemeHouse/XPress] => 1010570
            [XF] => 2021370
            [XFI] => 1050270
        )

    [wordpress] => /var/www/html
)

Intel is trapped - flawed vision, bad execution

Jumper

Member
The inspiration for this thread was this article talking about innovation traps. The commitment trap and business model trap really reminds me of intel. Let me explain.

Business-model trap:
Intel wants to transform to IDM 2.0 - sort of the factory for their own products and also products from other companies. Intel is investing a lot of money into factories and new staff but I heard they will have lots of problems with procuring enough EUV machines for those factories. Will they be able to sort this problem till they build these factories or will the factories just run o 75% capacity for example? There is also a problem with Intel nodes. Intel needs to fulfill its new roadmap(20A in 2024) to be competitive with TSMC. Will they be able to do it? Or are they just burning money trying to fight a competitor that has a clear edge over them?

Commitment trap:
I am talking mainly about gaming GPUs. Why are they pouring money into this? Alchemist is clearly not competitive and they could've spent that money on researching new nodes or they could've instead focused on fixing the server products development first. This would be a good idea if they had most of their areas under control but this seems like another black hole that is just steering their focus from important problems - bringing sapphire rapids to market, fulfilling their node roadmap.

Will Intel be able to focus with so many things on their plates or will the vision of Pat Gelsinger and their problems in execution lead them further into irrelevance?

Bonus question: What do you think about the new products from Intel? Will they be competitive against Nvidia and AMD?
 
I think that if intel wants to keep it's fabs, then IDM2.0 is a necessity. With the move to heterogeneous architectures Intel with have lower demand for it's newest nodes. Additionally as marketshare continues to slip Intel will need someone to buy their wafers. Entering the foundry business allows for this demand issue to be dealt with. This lower demand for new nodes can also help temper the EUV situation in the short/medium term, but as you pointed out there is definitely a risk that some of these new fabs might be partially idle for a bit until they get their EUV machines. As for the nodes, if intel can hit 20A in 2024 then they *should* have an advantage over TSMC. However if these goals slip maybe a year then I don't think that is the death of IFS. Because that should put them on par or ahead of Samsung. What will influence IFS's success far more is how easy it is to work with/design for intel, if their nodes provide value, and if Intel's timelines become semi-reliable. Intel preforming all of their fab work in the west should also slightly improve their demand with western fabless companies. Additionally if intel can provide nice perks (packaging, X86, design work, Tower IP, etc), that will also help them draw in more customers even if their nodes aren't 2-3 generations ahead of TSMC like they used to be.

As for the GPUs I think it is to be expected that the first generation is underwhelming. After all Radeon is only just now starting to become a powerful brand in PC/is probably the better part of a decade behind NVIDIA in the Datacenter/enterprise GPU software world. Going forward I think that GPUs are a big deal. Data centers want simplicity with their farms, and being able to have a bundled CPU/GPU blade is powerful. If intel can't get GPUs (to complement their other accelerators/CPUs) then they will be at a big disadvantage to AMD and NVIDIA. Furthermore the projected growth of AI and the datacenters that power both them and non AI workloads is crazy. Ignoring this new area might well be worse for intel than getting booted out of the phone/radio business by Qualcomm.

I do also fear that Intel might be overextended, but time will tell. One thing to keep in mind is that money spent on one can't necessarily be spent on other areas, and the sale of Intel's nand and 5G businesses might free up some resources for other areas.
 
I think that if intel wants to keep it's fabs, then IDM2.0 is a necessity. With the move to heterogeneous architectures Intel with have lower demand for it's newest nodes. Additionally as marketshare continues to slip Intel will need someone to buy their wafers. Entering the foundry business allows for this demand issue to be dealt with. This lower demand for new nodes can also help temper the EUV situation in the short/medium term, but as you pointed out there is definitely a risk that some of these new fabs might be partially idle for a bit until they get their EUV machines. As for the nodes, if intel can hit 20A in 2024 then they *should* have an advantage over TSMC. However if these goals slip maybe a year then I don't think that is the death of IFS. Because that should put them on par or ahead of Samsung. What will influence IFS's success far more is how easy it is to work with/design for intel, if their nodes provide value, and if Intel's timelines become semi-reliable. Intel preforming all of their fab work in the west should also slightly improve their demand with western fabless companies. Additionally if intel can provide nice perks (packaging, X86, design work, Tower IP, etc), that will also help them draw in more customers even if their nodes aren't 2-3 generations ahead of TSMC like they used to be.

As for the GPUs I think it is to be expected that the first generation is underwhelming. After all Radeon is only just now starting to become a powerful brand in PC/is probably the better part of a decade behind NVIDIA in the Datacenter/enterprise GPU software world. Going forward I think that GPUs are a big deal. Data centers want simplicity with their farms, and being able to have a bundled CPU/GPU blade is powerful. If intel can't get GPUs (to complement their other accelerators/CPUs) then they will be at a big disadvantage to AMD and NVIDIA. Furthermore the projected growth of AI and the datacenters that power both them and non AI workloads is crazy. Ignoring this new area might well be worse for intel than getting booted out of the phone/radio business by Qualcomm.

I do also fear that Intel might be overextended, but time will tell. One thing to keep in mind is that money spent on one can't necessarily be spent on other areas, and the sale of Intel's nand and 5G businesses might free up some resources for other areas.

Improving Intel's manufacturing technology is very important. But Intel is facing an even more urgent problem: Intel doesn't have new products that market needs. Those new products might or might not be manufactured with leading edge semiconductor manufacturing process.

If Intel can't come out new products that will generate significant amount of revenue, Intel regaining manufacturing leadership in 2025 will have limited effects.
 
I think that if intel wants to keep it's fabs, then IDM2.0 is a necessity. With the move to heterogeneous architectures Intel with have lower demand for it's newest nodes. Additionally as marketshare continues to slip Intel will need someone to buy their wafers. Entering the foundry business allows for this demand issue to be dealt with. This lower demand for new nodes can also help temper the EUV situation in the short/medium term, but as you pointed out there is definitely a risk that some of these new fabs might be partially idle for a bit until they get their EUV machines. As for the nodes, if intel can hit 20A in 2024 then they *should* have an advantage over TSMC. However if these goals slip maybe a year then I don't think that is the death of IFS. Because that should put them on par or ahead of Samsung. What will influence IFS's success far more is how easy it is to work with/design for intel, if their nodes provide value, and if Intel's timelines become semi-reliable. Intel preforming all of their fab work in the west should also slightly improve their demand with western fabless companies. Additionally if intel can provide nice perks (packaging, X86, design work, Tower IP, etc), that will also help them draw in more customers even if their nodes aren't 2-3 generations ahead of TSMC like they used to be.

As for the GPUs I think it is to be expected that the first generation is underwhelming. After all Radeon is only just now starting to become a powerful brand in PC/is probably the better part of a decade behind NVIDIA in the Datacenter/enterprise GPU software world. Going forward I think that GPUs are a big deal. Data centers want simplicity with their farms, and being able to have a bundled CPU/GPU blade is powerful. If intel can't get GPUs (to complement their other accelerators/CPUs) then they will be at a big disadvantage to AMD and NVIDIA. Furthermore the projected growth of AI and the datacenters that power both them and non AI workloads is crazy. Ignoring this new area might well be worse for intel than getting booted out of the phone/radio business by Qualcomm.

I do also fear that Intel might be overextended, but time will tell. One thing to keep in mind is that money spent on one can't necessarily be spent on other areas, and the sale of Intel's nand and 5G businesses might free up some resources for other areas.
Is Intel selling it's 5G business ? Hadn't heard that before, but I may have missed the news.
 
Is Intel selling it's 5G business ? Hadn't heard that before, but I may have missed the news.

Intel sold its 5G modem division to Apple in 2019 for US$1 billion. Some current Intel 5G related components are sourced from MediaTek.
 
Intel sold its 5G modem division to Apple in 2019 for US$1 billion. Some current Intel 5G related components are sourced from MediaTek.
OK, that part I knew. But that's only part of the Intel 5G business. Last I heard they still had a 5G base station/infrastructure business which they were still actively talking about in their investment calls. To the best of my knowledge they still have it (though there's much less talk about it these days).
 
Improving Intel's manufacturing technology is very important. But Intel is facing an even more urgent problem: Intel doesn't have new products that market needs. Those new products might or might not be manufactured with leading edge semiconductor manufacturing process.

If Intel can't come out new products that will generate significant amount of revenue, Intel regaining manufacturing leadership in 2025 will have limited effects.
This is part of why I think IDM 2.0 is critical. With how expensive(both to develop and to produce)/risky new nodes are, I don't think Intel can get their value out by the time it is time to switch. But if Intel can gather it's own niche, then they can squeeze more money out of these ever more expensive nodes while providing a lifeline if the IP side of things tanks. One of the big holes I see is that even their cheapest node is the relatively modern 22nm. And as we know with TSMC the real foundry money comes from the old stuff that gets produced for effectively free. Maybe this isn't that big of a deal due to the generally stagnate demand for these nodes, and maybe the tower deal helps sure up this weakness. The way I see it, IFS is a little too top heavy/specialized towards logic/HPC for my liking, but I would love to hear everyone's thoughts on the matter.
 
IDM 2.0 is brilliant and necessary for Intel. Most customers use the older nodes and we know that with intel they had a lot of 22 nm, then jumped to 14nm and were on 14 nm for a long, long while. They had to boost output production of 14nm due to the number of products they were making on that node. Also due to how long the were on that node it became very productive. Great yields, great speeds. Also Intel has bought out some big names like the FPGA maker Altera. Intels 5 nodes in 4 years means that they will earn more money back with other customers using those nodes, the customers benefit too. it really is a win/win for Intel. Helps pay for the expensive fabs worthwhile, and sure it will take a little while for Intel to step up in the fab making game due to ASML and tooling delays. There's no room in Intels life for share buy-back programs anymore. What a waste of money that was.
 
IDM 2.0 is brilliant and necessary for Intel. Most customers use the older nodes and we know that with intel they had a lot of 22 nm, then jumped to 14nm and were on 14 nm for a long, long while. They had to boost output production of 14nm due to the number of products they were making on that node. Also due to how long the were on that node it became very productive. Great yields, great speeds. Also Intel has bought out some big names like the FPGA maker Altera. Intels 5 nodes in 4 years means that they will earn more money back with other customers using those nodes, the customers benefit too. it really is a win/win for Intel. Helps pay for the expensive fabs worthwhile, and sure it will take a little while for Intel to step up in the fab making game due to ASML and tooling delays. There's no room in Intels life for share buy-back programs anymore. What a waste of money that was.
14nm is still fairly expensive due to the extensive use of double patterning. But I suppose that is why intel 16 is a relaxed 14nm, that way they don't need to do as much double patterning.
 
Intel 16 is basically 22FFL.
22FFL is 14nm transistors with fin and gate dimensions scaled back to be in between 14nm and 22nm and a simplified metal stack. If the data is to be believed, then the 16HP library can support drive currents that are closer 14nm than I thought would be possible given the dimensions. But the star of the show is probably the LL library, and I assume is what mediatech was interested in.

https://fuse.wikichip.org/news/567/...foundry-customers-targets-mobile-and-rf-apps/
 
Intel will be decimated again and again until they find their way of working with the fabless ecosystem.
 
Back
Top